Optimization of Detection and Calculation of Overlapping Mango Fruits and Illumination Variations Using Yolov11-Seg with Unsharp Masking Algorithm, Gamma Correction

Abstract

This study addresses the challenges of mango fruit detection and counting in complex horticultural environments, with a particular focus on the problems of occlusion (overlap) and extreme illumination variation in overhead drone imagery. To overcome the limitations of standard detection models in low-light conditions and for closely spaced fruits, we propose an integrated framework that combines YOLOv11-segmentation with advanced image enhancement techniques, namely Gamma Correction (GC) and Unsharp Masking (UM). Gamma Correction was used to normalize the light distribution in shadow areas, while Unsharp Masking supported by Bilateral Filtering enhanced edge contrast to distinguish mangoes from the dense foliage of the canopy. The methodology involved circular drone scanning at a constant distance of of 1.5 m to capture a diverse dataset in Maros Regency, Indonesia. The experimental results showed that the improved YOLOv11-segmentation model significantly outperformed the baseline model, achieving a mean Average Precision (mAP50) of 95.5%, an F1-score of 93.45%, and successfully identifying 2,085 true positive instances. This integration of pixel-level segmentation and preprocessing enabled the system to detect hidden fruits that the standard model failed to recognize, thus providing a robust and precise automated solution for crop management and yield estimation in the horticulture industry.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint